Chris -
I have a number of questions - but in order to properly frame them I need to
recount some history - please bear with me.
In the very early days of SR there was a proposal to reserve a fixed label
range on all routers (16000 - 23999) for the use of SR. This seemed workable
but there was concern that the same range would not be available on all routers
so the SR community agreed that we needed to allow the advertisement of a
different SRGB range on each SR capable router.
Then, another concern was raised that a single contiguous range large enough to
meet the SR requirements might not be available on all routers and therefore we
needed to support the advertisement of multiple SRGB ranges from a given node.
This capability has been added - though it has also raised other concerns
regarding how many ranges might be enough and how to deal with inconsistencies
(overlaps) in the advertised ranges should some error occur.
Now, in support of multiple algorithms/topologies, this draft proposes that
algorithm/topology specific SRGBs be advertised - and that to fully get the
benefits of consistent SID assignment for the same prefix across
algorithms/topologies (the major goal of this draft) that all routers be able
to support identical ranges for each algorithm/topology. (Section 5 goes into
this in detail) - and to be able to allocate new ranges for newly configured
algorithms/topologies on demand. Section 6 goes on to suggest that it would be
a wonderful goal if all routers could assign the same base label/range to each
algorithm/topology so that the labels could be derived algorithmically (though
you admit this may not always be possible).
My first question is how do you reconcile this requirement for consistency on
all nodes with the earlier decisions to support inconsistency as regards SRGB
base values/range sizes?
My second question is associated with the assumption of a single "node_index"
per node. I for one would be delighted if we could get agreement on using a
single host address/node (or at least one per AF) for all traffic to that node.
But this does not meet the deployment requirements in all cases and so we
support multiple "node_indeces" for a given node. This means that we do not map
SIDs to nodes - we map SIDs to prefixes. So it is conceptually more
straightforward and less problematic to configure SIDs for prefix ranges. This
is in fact how SRMS advertisements are constructed i.e.
(Prefix/Prefix length, Starting SID, Range)
How would you propose to incorporate this style of SID assignment into your
algorithms? Do you assume a maximum number of node-sids required for every node
in the network and require the range for each algorithm/topology be sized large
enough for the worst case? This seems potentially wasteful - especially if a
deployment chooses to use a particular address for a topology as a traffic
differentiator.
My third question relates to the algorithm/topology independence of the SID
assignment that you propose. Throughout the document you assume that you have a
single SID for a given prefix and you simply add the appropriate
algorithm/topology specific SRGB base value to this SID to derive the
forwarding label. However, current protocol advertisements for SIDs include
both algorithm and topology as context for the SID. This reflects the current
SR architecture which has been defined with algorithm/topology as an explicit
context for a SID. Your proposal fundamentally changes this and therefore all
existing encodings of SID advertisements (prefix reachability, SRGB, SRMS
advertisements) would have to be changed in a non-backwards compatible way in
order to support your proposal. Other than definition of a new SRGB
advertisement, you do not discuss this in the draft. How do you intend to
address this major issue?
Les
> -----Original Message-----
> From: spring [mailto:[email protected]] On Behalf Of Chris Bowers
> Sent: Monday, October 05, 2015 12:31 PM
> To: [email protected]
> Subject: [spring] FW: New Version Notification for draft-bowers-spring-adv-
> per-algorithm-label-blocks-02.txt
>
> SPRING WG,
>
> There was quite a lot of discussion in person after we presented draft-
> bowers-spring-adv-per-algorithm-label-blocks-01 in Prague, and there was
> also much discussion on this list about the issue raised by this draft. We
> have
> made two main additions to this revision of the draft to further this
> discussion.
>
> 1) We have included discussion of multiple topologies in addition to multiple
> algorithms, and we have modified the proposed ISIS extension to handle
> both multiple topologies and multiple algorithms.
>
> 2) We have tried to accurately describe a proposal, which was outlined on
> this list, for managing the assignment of per-topology/per-algorithm node
> index values, which we refer to as the "configured offset mapping method".
>
> We welcome feedback from the working group on this draft.
>
> Thanks,
> Chris
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]]
> Sent: Monday, October 05, 2015 2:05 PM
> To: Uma Chunduri <[email protected]>; Hannes Gredler
> <[email protected]>; Pushpasis Sarkar <[email protected]>; Chris
> Bowers <[email protected]>; Chris Bowers <[email protected]>;
> Uma Chunduri <[email protected]>; Pushpasis Sarkar
> <[email protected]>; Hannes Gredler <[email protected]>
> Subject: New Version Notification for draft-bowers-spring-adv-per-
> algorithm-label-blocks-02.txt
>
>
> A new version of I-D, draft-bowers-spring-adv-per-algorithm-label-blocks-
> 02.txt
> has been successfully submitted by Chris Bowers and posted to the IETF
> repository.
>
> Name: draft-bowers-spring-adv-per-algorithm-label-blocks
> Revision: 02
> Title: Advertising Per-Topology and Per-Algorithm Label Blocks
> Document date: 2015-10-05
> Group: Individual Submission
> Pages: 14
> URL: https://www.ietf.org/internet-drafts/draft-bowers-spring-adv-
> per-algorithm-label-blocks-02.txt
> Status: https://datatracker.ietf.org/doc/draft-bowers-spring-adv-per-
> algorithm-label-blocks/
> Htmlized: https://tools.ietf.org/html/draft-bowers-spring-adv-per-
> algorithm-label-blocks-02
> Diff: https://www.ietf.org/rfcdiff?url2=draft-bowers-spring-adv-per-
> algorithm-label-blocks-02
>
> Abstract:
> When segment routing is used in a network that is controlled by a
> link state IGP (such as ISIS or OSPF), each node in the network can
> be assigned one or more index numbers, known as "node-SIDs". The
> node-SIDs are unique within the network, and are known to all the
> nodes in the network. If an ingress node has a data packet to be
> sent to an egress node, the ingress node may select a node-SID
> corresponding to the egress node, and "translate" that node-SID to an
> MPLS label. The MPLS label represents a particular path to the
> egress node; the path is determined by applying a routing algorithm
> to a particular view of the network topology and a particular set of
> metric assignments to the links of that topology. The packet can
> then be forwarded by pushing the label on the packet's label stack
> and transmitting the packet to the next hop on the corresponding path
> to the egress node. This document compares two different procedures
> for translating a node-SID to the MPLS label that represents a path
> chosen by a particular algorithm operating on a particular topology.
> It also specifies the ISIS extensions needed to support one of the
> procedures (known as the "per-topology/per-algorithm label block"
> procedure).
>
>
>
>
> Please note that it may take a couple of minutes from the time of submission
> until the htmlized version and diff are available at tools.ietf.org.
>
> The IETF Secretariat
>
> _______________________________________________
> spring mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/spring
_______________________________________________
spring mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/spring
